Article Text

Download PDFPDF

Extended report
Continuous long-term anti-TNF therapy does not lead to an increase in the rate of new bone formation over 8 years in patients with ankylosing spondylitis
  1. Xenofon Baraliakos1,
  2. Hildrun Haibel2,
  3. Joachim Listing3,
  4. Joachim Sieper3,
  5. Jürgen Braun1
  1. 1Rheumazentrum Ruhrgebiet, Ruhr-University Bochum, Herne, Germany
  2. 2Department of Rheumatology, Charité University, Campus Benjamin Franklin, Berlin, Germany
  3. 3Epidemiology Unit, German Rheumatism Research Center, Berlin, Germany
  1. Correspondence to Dr Xenofon Baraliakos, Rheumazentrum Ruhrgebiet, Ruhr-University Bochum, Landgrafenstr. 15, 44652 Herne, Germany; baraliakos{at}


Objective Compare the radiographic progression of ankylosing spondylitis (AS) patients treated with infliximab (INF) versus historical controls (Herne cohort, HC) never treated with tumour necrosis factor (TNF)-blockers over 8 years.

Methods Patients were selected based on the availability of lateral cervical and lumbar radiographs at baseline (BL) and after 8 years. Radiographs were scored by two blinded readers using modified Stokes AS spinal score (mSASSS). Mixed linear models were applied to compare radiographic progression between cohorts after adjustment for baseline status.

Results Patients in INF (n=22) and HC (n=34) did not differ in the mSASSS status: 13.2±17.6 in INF versus 14.2±13.8 in HC (p=0.254). Both showed progression at 8 years: mean mSASSS 20.2±21.4 in INF and 25.9±17.8 in HC. After adjustment for baseline damage the mean mSASSS (SEM) at 8 years was 21.0 (1.4) in INF and 25.5 (1.1) HC (p=0.047). The mean mSASSS difference was similar in the groups between baseline and 4 years but was more pronounced in HC between 4 and 8 years (p=0.03 between groups). The mean number of syndesmophytes, although similar at baseline, differed significantly at 8 years: 1.0±0.6 new syndesmophytes/patient in INF versus 2.7±0.8 in HC (p=0.007). Adjustment for age, symptom duration, HLA-B27, Bath AS disease activity index and Bath AS function index at baseline had no influence.

Conclusions Despite limitations of patient numbers and retrospective study design, these data show increase in new bone formation in both patients treated with anti-TNF and those who did not. However, since there was even less bone formation in the INF treated group after 8 years, these data argue against a major role for the TNF-brake hypothesis.

  • Ankylosing Spondylitis
  • Anti-TNF
  • NSAIDs

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


Ankylosing spondylitis (AS) is a frequent chronic inflammatory rheumatic disease which affects mainly the axial skeleton of patients in the second and third decades of life.1 It starts in the sacroiliac joints and spreads to the spine in later stages.2 The main clinical complain is inflammatory back pain. Characteristic features for the inflammatory nature of AS are sacroiliitis and spondylitis. MRI is considered the standard imaging technique in the detection of such lesions3 but conventional radiographs are still the gold standard for assessment of structural changes, such as new bone formation, which are pathognomonic for the diagnosis of AS4 and have been included in the internationally accepted Assessments of Spondyloarthritis international society (ASAS) core set for AS.5

Conventional treatment of AS with non-steroidal anti-inflammatory agents (NSAIDs) and physiotherapy is the standard of care, while treatment with tumour necrosis factor α (TNFα)-blockers is the only alternative for AS patients who remain active despite NSAID therapy.6

While inflammatory lesions as detected by MRI which were reported to be linked to new bone formation in AS7–9 may improve in patients treated with TNFα-blockers already within a short time period,10 such treatment has not shown benefit regarding the progression of structural changes after 2 and 4 years of continuous treatment when compared with historical cohorts.11–15 On the other hand, although no effect could be seen on inflammatory spinal changes in a small study,16 NSAIDs were shown to reduce the rate of radiographic progression in AS patients, especially in those with high serum levels of C reactive protein (CRP) either when given continuously or in higher dosages.17 ,18 Whether this effect of NSAIDs is also relevant in patients undergoing anti-TNF therapy is unknown.

The quantification of chronic spinal changes assessed by conventional spinal x-rays in AS is usually done with the modified Stokes AS spinal score (mSASSS) which the best evaluated method,19 ,20 even though there has been some discussion regarding the inclusion of both erosive and osteoproliferative changes and the lack of assessment of the thoracic spine.21

In patients with AS, treatment with TNFα-blockers may be needed for long periods of time since attempts to withdraw this therapy have not been very successful.22 ,23 Furthermore, both the important patient reported outcome function (Bath AS function index, BASFI24) and the physician assessed metrology (Bath AS metrology index25) have shown sustained long-term improvement under TNFα-blocker therapy.26 Nevertheless, the question of a possible disease modifying effect of anti-TNF therapy on radiographic progression, for example, new bone formation, over longer periods of time is of major clinical relevance.27 However, longer-term follow-up data exceeding 2 years from patients treated with TNF-blockers are scarce.

Furthermore, recent data have shown that syndesmophytes are more likely to develop after resolution of inflammation28 and in areas of fatty degeneration29 in patients treated with TNFα-blockers. The so-called TNF-brake hypothesis is based on the connection of TNFα to the Wnt pathway;8 it suggests that blocking TNF may rather stimulate bone formation, and indeed bone mineral density has been shown to substantially improve in patients treated with infliximab.30 However, whether this mechanism is also relevant for syndesmophyte formation is unknown.

Taken together, an investigation of the long-term effects of TNFα-blockers on the radiographic progression in patients with AS is mandatory. In this study, we took the spinal radiographs available from the first investigator driven multicentre study with the TNFα-blocker INF in patients with AS31—currently the only study with an extension for up to 8 years (‘Deutsche Infliximab Kohorte für AS, ‘DIKAS’,26 and compared them with 8-year follow-up x-rays of patients collected in our historical AS cohort (Herne cohort, ‘HC’32)) of patients who had never been treated with TNFα-blockers.


Patient groups and radiographic assessment

A total of 56 patients were included in this study, all of whom fulfilled the modified New York diagnostic criteria for AS.33 Overall, 22 patients were participants in DIKAS (initially included 69 patients) and 34 patients were taken from HC (initially included 146 patients). In DIKAS, all patients were treated according to a standard protocol with 5 mg/kg INF intravenous continuously every 6 weeks. Patients from HC were all hospitalised in our clinic between 1993 and 2005 for different reasons and were not treated with TNF-blockers either because these were not available at that time or due to personal reasons of the patients. As described earlier, reasons for admission are high levels of pain, increased disease activity and also disability and peripheral symptomatic inflammation associated with functional decline. All patients were treated conventionally with NSAIDs and none of them had ever received any additional TNFα-blocker.

The selection of patients was made according to their availability of conventional radiographs of the cervical and the lumbar spine at baseline (BL) and 8 years; most of them also had radiographs at 2 years (n=35), 4 years (n=33) and 6 years (n=44), respectively.

After blinding for patients’ treatment and time point of imaging, all images were scored by two experienced readers (XB and HH) using the mSASSS.19 The use of the mSASSS for the assessment of structural changes in AS has been described in detail elsewhere.19 ,34 Briefly, it evaluates the anterior part of the lumbar and the cervical spine by assessing chronic changes using a score between 0 and 3. The score ranges from 0 to 72 scoring points for both spinal segments. Since not all images may appear in perfect quality (overexposure or underexposure of the radiograph) or that the spinal segments are not always completely captured on one film, which might lead to missing of some sites, we excluded patients who had more than three vertebral sites missing. In the cases with ≤3 vertebral sites missing, the missing scores were substituted by the mean score of the vertebra of the same spinal segment of the patient. Both of these procedures have been proposed34 and were used in several AS studies in the past.

For the analysis, the mean mSASSS scores of both readers were used. In addition, the mean numbers of syndesmophytes per patient was calculated for each patient group. For this analysis, only those vertebral edges where both readers showed agreement on the occurrence of a syndesmophyte were taken into account.

Statistical analysis

The Mann–Whitney test was used for simple comparisons between both cohorts. A mixed model approach was used to compare radiographic progression between both cohorts by taking the baseline status, follow-up time and the interaction between follow-up time and treatment regime (NSAIDs (HC) vs INF (DIKAS)) into account. As possible risk factors the following baseline characteristics were considered: mSASSS, age, symptom duration, human leukocyte antigen B27 (HLA-B27), Bath AS disease activity index (BASDAI) and BASFI. Since during calculations only an association with the mSASSS at baseline was found to be significant, only the mSASSS status at baseline was used for adjustment in the final model. As outcome parameter, the mean mSASSS scores of both readers were used in the mixed model analysis. Since in a mixed model analysis an adjustment for missing data is made by a statistical formula, a replacement of missing data was not performed in this analysis. However, since mSASSS scores are skew distributed, the mixed model was also applied to square root transformed mSASSS scores in a sensitivity analysis. Furthermore, to investigate whether the mean changes in the mSASSS scores were confounded by only a small number of extreme values, missing data at 4 years were replaced by imputation methods (SAS procedure MI) and individual changes were then visualised by probability plots. The results shown in these plots were compared by an analysis of covariance based on rank transformed data with the mSASSS baseline status as covariable. A validated SAS macro was used for the calculation ( Furthermore, a non-parametric test was used to investigate whether patients with missing 8 years mSASSS data differed in their radiographic progression from patients with complete data included in this study.35

‘Agreement’ between readers was defined as no difference in the same mSASSS change scores between time points. ‘Some disagreement’ was defined as a small difference of ≤2 mSASSS units and ‘major disagreement’ was defined as disagreement >2 mSASSS units of change (development of new syndesmophyte/ankylosis, see below) between two assessment time points, as described recently.32 In case of major disagreement discrepancies between two readers (difference >2 mSASSS units,), a senior reader (JB) had to re-evaluate the image and give the final score. Reliability and agreement between readers was made by comparison of individual films and analysis of agreement, some disagreement and major disagreement. Statistical analysis was done by using SAS V.9.3.


Patient's characteristics at baseline

The patients included in this study had similar baseline characteristics compared with patients included in the original cohorts.31 ,32

The baseline data and differences between cohorts are presented in table 1. Patients in DIKAS were younger and had shorter mean symptom duration. Furthermore, patients in DIKAS had higher BASDAI and BASFI levels but the mean BASDAI in HC was also ≥4 units.

Table 1

Baseline characteristics of both groups used for the 8-year radiographic comparison

The comparison of mSASSS progression in patients who were excluded from this analysis due to incomplete sets of radiographs versus patients who were included due to radiographs available at year 8 showed a significant difference in HC (10.6±2.6 vs 5.8±4.3, respectively, p=0.0026), whereas in DIKAS this difference was not significant (3.1±2.8 vs 5.0±5.3, p=0.068).

Analysis of the radiographic progression over 8 years as assessed by the mSASSS

The degree of radiographic damage at baseline was similar in both groups (13.2±17.6 in DIKAS vs 14.2±13.8 in HC, p=0.26).

Overall, both groups showed significant radiographic progression after 8 years, with an mSASSS of 20.2±21.4 in DIKAS and 25.9±17.8 in HC (both p<0.001 as compared with BL). Furthermore, the mean radiographic progression per year was 0.9±0.8 in DIKAS and 1.5±1.4 in HC (p=0.129).

After adjusting for baseline mSASSS (and, indirectly, for missing values), a significant difference of the radiographic progression between the groups was found (figure 1, p<0.0001 for the overall difference (interaction between follow-up time and cohort)). Other baseline parameters (age (p=0.61), symptom duration (p=0.42), HLA-B27 (p=0.10), BASDAI (p=0.53) and BASFI (p=0.38) at BL gender (p=0.93)) were not found to be significantly associated as confounders for comparison of the radiographic progression between the two groups. At 8 years, the mSASSS difference between the HC and DIKAS was 4.5 units (figure 1, p=0.047 on normalised (square root transformed) mSASSS data). Importantly, the difference in the mean scores was not dependent on single extreme values (figure 2A,B). Radiographic progression between years 4 and 8 differed significantly between both treatment groups (Analysis of covariance based on ranks, p=0.01, figure 2B), whereas no significant differences were observed for the first 4 years of comparison (figure 2A, p=0.18).

Figure 1

Radiographic progression of the two treatment groups. aTNF: patients in DIKAS (Deutsche Infliximab (INF) Kohorte für ankylosing spondylitis), treated continuously with INF, HC: patients in the Herne Cohort, treated continuously with non-steroidal anti-inflammatory agents, no biologic treatment was applied to these patients. Values are given as means adjusted for baseline and missing data at follow-up, SE of means in brackets and 95% CIs as error bars.

Figure 2

Cumulative probability plot for the changes in modified Stokes ankylosing spondylitis spinal score (mSASSS) of both assessed cohorts during the first 4 years (A) and between years 4 and 8 (B) of continuous treatment with infliximab (INF) (DIKAS (Deutsche INF Kohorte für AS)) and non-steroidal anti-inflammatory agents (Herne cohort (HC)). Each single sign (circle or triangle) represents the mSASSS change in one individual patient. During the first 4 years (A) no differences were seen in the radiographic changes of the patients between the two treatment groups, while in the longer-term follow-up (year 4–year 8, B), patients in the HC showed higher rates of radiographic progression in the mean mSASSS (see also figure 1) and on the individual level.

Analysis of the radiographic progression over 8 years as assessed by the development of new syndesmophytes

The amount of baseline syndesmophytes was similar in both groups with 3.6±5.8 in DIKAS and 3.7±4.0 in HC (table 2). After 8 years, patients in HC had significantly higher numbers of syndesmophytes (6.4±4.8) as compared with patients in DIKAS (4.6±6.4, p=0.007) (table 2). Furthermore, the mean amount of new syndesmophytes increased significantly only in HC, while in DIKAS there was a trend but not a significant increase (table 2).

Table 2

Mean amounts of syndesmophytes at baseline and after 8 years of treatment in both groups

In more detail, there were differences in the amount of patients with versus without baseline syndesmophytes between the two different groups: 12/22 patients (54.5%) in DIKAS and 27/34 (79.4%) in HC (p=0.157) had baseline syndesmophytes while 10/22 patients (45.5%) in DIKAS and 7/34 (20.6%) in HC (p=0.225) had no baseline syndesmophytes. After 8 years, patients without baseline syndesmophytes in HC showed a trend for more new syndesmophytes (2.6±4.7 new syndesmophytes per patient) versus patients in DIKAS (0.8±1.6 new syndesmophytes per patient), p=0.36. For patients without baseline syndesmophytes, this difference was significant with less new syndesmophytes in DIKAS (1.3±4.5 new syndesmophytes per patient) versus patients in HC (3.3±1.9 new syndesmophytes per patient), p=0.032.

Reliability and agreement between readers

Both readers were in agreement in the change scores of the mSASSS in 35/56 patients (62.5%), while there was some disagreement in 18/56 patients (32.1%) and major disagreement in 3/56 patients (5.4%). The agreement was higher in those patients who had an mSASSS score of ‘0’ at baseline (as scored by both readers, data not shown).


This study on the long-term effects of anti-TNF therapy is the first to analyse new bone formation over 8 years of patients with AS. Similar to previous studies we used a historical cohort that had not been treated with TNF-blockers in which spinal radiographs were available over a similar time period. The data of our study suggest that although continuous anti-TNF therapy is associated with new bone formation, the rate of new bone formation over time is not increased in comparison with a patient group that had not been treated with anti-TNF agents.

The images and the data used in this study were in part from our investigator driven clinical trial,31 which was accepted by the European Medicine Agency as the major data source for the approval of INF for the treatment of patients with active AS some years ago. This study has been extended for a 10-year follow-up and some patients had agreed to have conventional x-rays performed at several time points. It was decided to use the 8-year follow-up data because after 10 years, even less radiographs were available.

For the quantification of radiographic spinal changes we used two different approaches: one approach was the use of the best evaluated x-ray scoring method available, the mSASSS,34 ,36 which assesses the lateral radiographs of the cervical and the lumbar spine with only the ventral part of the vertebral bodies being scored. The second approach was to count the number of new syndesmophytes developed over time, which is the more practical way of assessment of radiographic progression and is also more interesting for clinicians in daily clinical practice.

Using the scoring system mSASSS, continuous radiographic progression was observed in both groups, for example, the patients under long-term treatment with a TNF-blocker and the historical cohort. Accordingly, the mSASSS score was significantly increased after 8 years, as compared with baseline in both groups.

When comparing the radiographic progression rates in both our cohorts with published data from other studies, the progression rate observed here seems relatively high. However, as stated above, to date only studies with 2-year data are available and these consistently showed an mSASSS progression of about 1 mSASSS unit within these 2-year periods. Theoretically, this would imply that in case of a linear mSASSS increase over time, a progression of 4 mSASSS units would have been expected. However, this may not be correct because there is evidence that in at least 25% of the patients the progression rate is clearly not linear32 and, in addition, it can be expected that the expected progression rate is not increasing linearly since every newly developed syndesmophyte has the potential to accelerate the expected progression rate.4 ,37 Finally, one needs also to keep in mind that the patients who have been included early in clinical trials are usually the most severe ones, adding even more weight to this accelerated progression with longer observation periods.

There was somewhat less bone formation in the INF group in comparison with the historical cohort.32 However, interestingly, although the mSASSS change in the first 4 years of comparison was similar it became different in the following 4 years. This finding is also of interest when comparing the baseline status of the cohorts, since the patients in DIKAS had a potentially worse outcome status at baseline, with higher disease activity (BASDAI, CRP) and higher functional limitations (BASFI). Furthermore, those patients in HC who were excluded from the analysis due to missing radiographs after 8 years were the ones who showed the highest progression rates. Importantly, similar results were also obtained when the number of newly developed syndesmophytes over 8 years was assessed. The mean amount of new syndesmophytes was only 1/patient in DIKAS versus almost 3/patient in HC. Again, there were similar differences in the baseline status of the patients, with more patients having baseline syndesmophytes in the DIKAS cohort.

Some limitations of this study have to be discussed: it is mainly the retrospective study design and the relatively low number of patients and available radiographs. Another limitation is that we could only collect radiographs of the patients who remained in the study and who may subsequently have better clinical outcomes. The patients who dropped out of the study have not been followed up more closely and radiographs were not collected. However, the reasons of discontinuation from the study were in many cases pragmatic and in general not due to inefficacy of the drug.26 Some of the patients who discontinued due to an adverse event might even have changed to another anti-TNF agent.38

Furthermore, our historical cohort may be a comparatively severe one. However, this applies to all comparisons with historical cohorts, which are of course not the same as direct comparisons. However, based on the well-documented clinical efficacy of the TNF-blockers also over longer periods of time,26 long-term placebo-controlled studies are generally not considered to be ethically feasible. Nevertheless, the observed rate of radiographic progression after 8 years of anti-TNF treatment, although clearly increasing over time (which confirms many earlier data),12–14 does not appear to relatively increase over 8 years. The radiographic outcome in the TNFα-blocker treated patient group with prevalent syndesmophytes at baseline is important in the background of the assumed pro-osteoblastic effect of this treatment.28 As suggested in this so-called ‘TNF-brake hypothesis’,8 TNF blockade may rather stimulate bone formation through continuous suppression of inflammation. However, our results indicate that this is rather unlikely since even the subgroup of patients at higher risk of radiographic progression did not have more newly developing syndesmophytes.

Recently, there is some evidence that NSAIDs have an effect on radiographic progression.17 ,18 Therefore, the assessment of NSAID intake39 would have been of interest but was not performed in this study because initially it was not planned. However, the patients of HC were usually treated with relatively high doses of NSAIDs over longer periods of time because they had been admitted to the hospital.32

Taken together, this long-term follow-up of patients being treated with TNFα-blockers over 8 years including the comparison with a historical cohort of patients suggests that radiographic progression in general and syndesmophyte formation in particular does progress in patients treated with TNFα-blockers. However, and this seems important regarding recent discussions,27 ,28 in comparison with this historical cohort the progression rate was clearly not higher and even lower. It remains to be seen whether there will be more studies with larger patient numbers to confirm these results. Hopefully, projects such as the European AS infliximab cohort40 ,41 will be even more successful in recruiting more patients for long-term analyses which are clearly needed to be able to make recommendations for the duration of anti-TNF therapy over longer periods of time.


Correction notice This article has been corrected since it was published Online First. The acknowledgements section has been updated.


We would like to thank all patients who participated in this study for their participation over the entire study period and their willingness to provide their images for study analyses. Furthermore, we would also like to thank the medical stuff of the all participating centers, especially those who contributed with continuous collection of the conventional radiographs throughout the study, in the Charité – Campus Benjamin Franklin (Prof. Dr. J. Sieper), University of Hannover (Prof. Dr. H. Zeidler), Rheumatology practice Dr. Soerensen (Dr. B. Andermann), University of Munich (Prof. Dr. Schewe/Dr. Grünke) and University of Düsseldorf (Prof. Dr. M. Schneider) and all study nurses who were involved in this study and made the coordination of the study possible, especially Beate Buss from Charité, Medical University of Berlin, Germany and Anna von der Recke, MD, Herne, Germany.



  • Handling editor Tore K Kvien

  • Contributors XB: idea, writing of manuscript, reading of radiographs. HH: review of manuscript, reading of radiographs. JL: statistical analysis, writing of manuscript. JS: idea, review of manuscript. JB: idea, writing of manuscript.

  • Funding There was no funding for the present study. MSD Germany served as funder for the study drug for the clinical part of the DIKAS study and had no influence on collection of study data or analyses at any time of the study period.

  • Competing interests None.

  • Patient consent Obtained.

  • Ethics approval The local independent ethics committees of all participating centers approved the study and the performance of the x-rays.

  • Provenance and peer review Not commissioned; externally peer reviewed.